Three frontier models in two days

On the first of September, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1. On the second, Google released Gemini 3.8 Flash and Meta released Muse Spark 1.3. Three frontier releases in the first two days of the month, and none of them a price cut on what came before.

If you are reading that as news, it is mildly interesting. If you are about to have a product built, it is the most important thing on this page.

Here is why. When someone builds you an app and wires it directly to whichever model was best the week they started, they have not given you a product. They have given you a bill that arrives every few months, because the thing underneath keeps moving and somebody has to keep up with it. That somebody is you, or someone you pay by the hour.

The alternative is not to pick the right model. There is no right model for very long. The alternative is to keep the model in a layer you never open. When a better one lands, it goes in there, and what your customers see gets better without you deciding anything or paying for a rebuild. That is an architecture decision, and it is made once, at the start, or it is not made at all.

The second thing that happened is quieter and matters more. In the last week of August, more than a hundred companies — OpenAI, Anthropic, Google, Microsoft, AWS, Cloudflare, CrowdStrike, Visa among them — signed an open letter called A Call for Collective Action on Cyber Defense. It warns that AI-enabled attacks are about to get much more common, and it asks the industry for three things: AI-powered defences, trusted model access, and traceable agent identities.

That last phrase is the one to hold on to. Traceable agent identities means being able to prove which automated thing did what, and when. Not a log somebody could write after the fact — an identity the system enforces while the work is happening.

You cannot bolt that onto a finished product. It is a decision about how the thing is built: every action attributable, every gate recorded, an agent that always says it is AI because European law requires it to. Built in at the start it costs nothing extra. Added in a year, under pressure, after something has gone wrong, it is a rewrite.

None of this is a reason to be nervous about building. It is a reason to be careful about what you personally take on. You know your niche and you can sell into it. Tracking four model launches in a week, and what a hundred companies are asking regulators for, is not a good use of the person who knows the customers.

That is the split. You run the business. We run the system — including the part of it that changed twice this week while you were selling.